Estimation of Tongue Motion and Vowels of Silent Speech Based on EMG from Suprahyoid Muscles using CNN

Abstract

In this paper, we propose a method to estimate the tongue motion direction and silent speech based on convolutional neural network (CNN) using the surface electromyogram (EMG) from the suprahyoid muscles. Conventional human machine interface (HMI) is difficult to use for users who are unable to freely move the muscles below the neck due to nerve damage or the like. Therefore, we have developed a method to estimate the tongue motion in 6 directions and 5 vowels of silent speeches from 4 channel EMG. As a result of verification experiment, we obtained averaged accuracy was about 81.2% in the estimation of the tongue directions and the silent speeches. Thus, it was suggested that simultaneous estimation is possible based on EMG measured from electrodes on the anterior neck region.